Application of some feature mapping methods in QNAR prediction of cellular uptake of magneto fluorescent nanoparticles

In this work, the cellular uptake of 109 magneto nanoparticles (MNPs) in human pancreatic cancer cells (Paca2) were predicted by applying quantitative nanostructureactivity relationship (QNAR) methodology. The most important descriptors selected by stepwise multiple linear regression (SW-MLR). Some feature mapping techniques such as random forest (RF), multiple linear regression (MLR) and least square support vector machine (LS-SVM) were used for development of QNAR model. Inspection to these models indicates LS-SVM model is finely capable for predicting the cellular uptake of MNPs. For this model, the correlation coefficient (R) was 0.935 and 0.933, and the root-mean square error (RMSE) was 0.16 and 0.23 for the training and test sets, respectively. The built LSSVM model was assessed by leave one out cross-validation (Q2= 0.53, SPRESS=0.16) as well as external validation. In addition, sensitivity analysis of LS-SVM model indicated the role of electronic and steric interactions of MNPʼs organic coating are the predominant factors responsible for cellular uptake in Paca2.